🎬 Gen-3 Best Video AI Yet

PLUS: Karpathy: Neural Nets to Replace Software

Welcome back!

The evolution of generative video is racing ahead at an insane pace. Runway latest Gen-3 Alpha model is now generally available and early users are amazed by its quality and realism. Let's unpack...

Today’s Summary:

  • Runway launches Gen-3 Alpha

  • Karpathy predicts neural nets will replace software

  • Meta new 3DGen AI model

  • Anthropic funds new model benchmarks

  • Figma AI disabled

  • YouTube addresses deepfakes

  • 3 new tools


AI Video Generation Reaches New Heights with Gen-3 Alpha

The Summary: Runway has launched Gen-3 Alpha, a powerful text-to-video AI model, now generally available. Initial user reactions are positive, with many impressed by the tool's capabilities. The model allows users to generate high-fidelity videos from text prompts with remarkable detail and control.

Users are particularly excited about Gen-3 Alpha improved quality and realism compared to all competitors. While currently limited to paid plans, the tool is already enabling creators to explore novel concepts and scenarios.

  • Text to Video (released), Image to Video and Video to Video (coming soon)

  • Offers fine-grained temporal control for complex scene changes and transitions

  • Major improvements in fidelity, consistency, and motion

  • Paid plans are currently prioritized, at $76/mo for unlimited generations, with rate restrictions. Free limited access should be available later

  • Prompting guide

  • RunwayML rumored to be in talks to raise $450M at $4B valuation

Why it matters: Gen-3 Alpha opens new creative possibilities for artists and content creators. Its ability to generate complex, detailed videos from text inputs could change video production workflows. Gen-3 Alpha stands out for its quality, surpassing recent models despite its higher access price. While it targets professionals more than the general public, it appears to be the top tool in the field since OpenAI's Sora, which remains still inaccessible.


Karpathy Envisions Neural Nets to Replace Software in the Future

The Summary: Andrej Karpathy, one of the most prominent figures in AI, predicts a future where computers will consist of a single neural network, with no classical software.

His vision includes devices that directly feed inputs like audio and video into the neural net, which then outputs directly to speakers and screens. Karpathy's statement has sparked discussions about the practicality and implications of such a radical shift in computing architecture.

Key details:

  • The proposed system would be "100% Fully Software 2.0"

  • Device inputs (audio, video, touch) would feed directly into the neural network

  • Outputs would be displayed as audio/video on speakers/screens

  • Some reactions express excitement, while others question practicality

  • Concerns raised include compute requirements and debugging challenges

Why it matters: This long-term prediction from a respected AI researcher suggests a fundamental change in how we might interact with computers in the future. If realized, it could lead to intuitive computing experiences, but also raise challenges in security and user control. The provocative idea pushes boundaries of computing paradigms, potentially influencing future research. It highlights the growing recognition that current OS architectures may not meet the demands of advanced AI applications.


Meta 3DGen Generates Complex 3D Models in Under a Minute

The Summary: Meta Research has introduced 3DGen, a new AI system that creates high-quality 3D assets from text prompts in less than a minute. 3DGen combines two powerful components: AssetGen for initial 3D generation and TextureGen for enhanced texturing.

The system outperforms leading industry solutions in prompt fidelity and visual quality, especially for complex scenes. 3DGen supports physically-based rendering, allowing generated assets to be used for real-world applications.

Key details:

  • Generates 3D assets with high-res textures and material maps

  • Produces results 3-10x faster than existing solutions

  • Supports PBR (physically-based rendering) for realistic lighting

  • Can generate new textures for existing 3D shapes

  • Outperforms baselines on prompt fidelity

  • Evaluated by professional 3D artists and general users

  • For now only research paper published, code still not released

Why it matters: This technology could be transformative for video games, AR/VR, and film industries. 3DGen may enable personalized and user-generated 3D content. The system speed and quality improvements could accelerate development cycles and open new possibilities for both professionals and hobbyists in the 3D and AR/VR space.


Quick news


🥇 New tools

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/